JPH05174005A - Method and device for character processing - Google Patents

Abstract

PURPOSE:To effectively decide the structures of paragraphs in the KANA (Japanese syllabary)/KANJI (Chinese character) conversion processing. CONSTITUTION:In a character processing method, where the KANA/KANJI conversion is carried out, an inputted KANA string is converted into a KANJI- KANA sentence based on a dictionary WDIC which stores the Japanese reading KANA of the words and the information on the words including the descriptions with correspondence secured between them. Then plural groups are defined to the word groups included in the dictionary WDIC, and the group likelihood BYUD showing the emerging easiness of the words belonging to the groups are stored for each group. Then the BYUD corresponding to at least one group is updated in response to the group where the words undergone the KANA/ KANJI conversion belong. In the KANA/KANJI conversion processing state, the BYUD corresponding to the groups where the words forming the candidates are compared with each other among the candidates of paragraph structures prepared to the KANA character strings to be converted. Thus the structure of the paragraph to be converted is decided.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character processing method and apparatus for efficiently determining a phrase to be converted and output from homonymous phrases in the kana-kanji conversion process.

[0002]

2. Description of the Related Art Conventionally, a Japanese word processor that creates a Japanese sentence using Kana-Kanji conversion for converting an input reading into a Kana-Kanji mixed sentence has been widely used.

However, Japanese has a characteristic that there are many words (homonyms) and phrases having the same reading but different meanings. In order to deal with this, the homonyms must be correctly selected according to the situation. The homonym synonym selection technology for conversion and output has become an important technology in kana-kanji conversion processing.

Conventionally, as a homonym selection technique, a short-term learning method in which a word used closer to the time when a sentence is created is prioritized as a candidate, and an example conversion method using co-occurrence between adjacent words have been adopted.

The short-term learning method is a method in which the most recently used word is used as a conversion result when determining a word to be converted from homophones, and has an advantage that the algorithm is simple and efficient. It has been used since ancient times. However, since the situation of the word string before and after the target homonym is not taken into consideration, there is a disadvantage that it does not function properly when another homonym is already used in the same sentence. For example, if you enter the sentence "Summer is hot." By Kana-Kanji conversion and confirm it, and then enter the reading "Autiochagana-like." Since the word "hot" is used, it will be converted into "I want to drink hot tea."

[0006] The example conversion is a technique for compensating for the shortcomings of short-term learning as described above. That is, if a set of words that co-occur with a semantic connection is used frequently,
The set of words is stored in advance, and if such a set of words matches when performing Kana-Kanji conversion, that set of words is used as the conversion result. For example, the two words “hot” and “tea” are a set of words that are often used in the form that “hot” adorns “tea”. Therefore,
Memorizing this set of words, and converting the reading "Atsuiocha" into kana-kanji, there are "hot", "thick", "hot", etc. as homonyms for "Atsu". “Hot” that matches the set of words that are present is determined as the conversion result and converted to “hot tea”. This can make up for some of the shortcomings of short-term learning.

However, as a drawback of the example conversion, there is a problem that it is completely useless for a word that appears without having a semantic connection. For example, if the word "shikou suru." Is entered when creating a sentence about law, there is no word that has a co-occurrence relationship with "shikou". Therefore, in the example conversion, the homonym "shikou" The word to be converted cannot be determined from among "thought", "oriented", and "execution".

Therefore, in order to solve this problem, a method using a word network has been proposed.
No. 167. Word network is 1
It is an interconnected network in which one word is associated with one node and each node is interconnected by a link having a value indicating a semantic distance between corresponding words. In addition, each node has a value that indicates how easily it appears in the document, the value of the corresponding node is increased each time a word is used, and the value of the increased node is linked. The value of the node that corresponds to a word that is propagated by and is semantically close becomes large.

In kana-kanji conversion, when determining a word to be converted from homonyms in the network, it is converted into a word having the highest value of the corresponding node. For example, since "law" and "enforcement" are semantically close to each other, the node corresponding to "law" and the node corresponding to "enforcement" are linked by a link having a large value, but "law" Since "thinking" and "oriented" are not close in meaning, the node corresponding to "law" and the node corresponding to "thought" and "oriented" are connected by a link having a small value. If the word "law" is used when creating a sentence about law using such a word network, the number of nodes corresponding to "law" is increased. While propagating through the network, the value of the node corresponding to "enforcement" linked by links with large values is also increased, but "thinking" and "oriented" linked by links with small values The value of the node corresponding to is not so high. Therefore, in this state, if the reading "Shikou suru." Is entered, the value of the corresponding node is the highest among the homonyms "thinking", "direction", and "enforcement". Therefore, "enforce." Is determined as the conversion result.

[0010]

However, the above-mentioned word network is a network in which one node is assigned to one word and the nodes are mutually connected by links. Since the link has a value, it has a drawback that it requires a huge memory capacity. For example, if 100,000 words are registered in the word dictionary, the number of nodes in the corresponding word network is 100,000, and the number of links is 10 billion.

Further, since the values of all the nodes have to be recalculated in order to propagate the value of the node corresponding to the used word to other nodes, there is a drawback that a huge amount of calculation is required. there were.

Further, the above network is intended to be used for selecting candidates in word units, and was not considered to be used for determining a bunsetsu structure including a delimiter position of a bunsetsu of a character string including a plurality of bunsetsu. ..

[0013]

In order to solve the above problems, according to the present invention, an input kana character string is stored in association with the reading of a word and the information of the word including notation. In the character processing method for executing a kana-kanji conversion process for converting a kanji-kana mixed sentence by referring to the dictionary, a plurality of groups are defined for the word group stored in the dictionary, and A group likelihood indicating the likelihood of appearance of a word belonging to the group is stored for each group, and the group likelihood corresponding to at least one of the groups is stored according to the group to which the word converted by the kana-kanji conversion process belongs. In the kana-kanji conversion processing, the kana-kanji conversion process corresponds to a group to which a word or phrase constituting the candidate belongs, out of candidates of the structure of the phrase for the kana character string to be converted. Compare the loop likelihood, and determines the structure of the phrase to be translated.

According to another aspect of the present invention, the character processing device includes a dictionary in which word readings and word information including notations are stored in association with each other, and input means for inputting characters. A kana-kanji conversion means for converting the kana character string input by the input means into a kanji-kana mixed sentence by referring to the dictionary; group information storage means for storing a group to which the word belongs, At least one of the groups is stored according to the group to which the word converted by the kana-kanji conversion means stores the likelihood of appearance of a word belonging to a group as a group likelihood for each group. Group likelihood updating means for updating the group likelihood corresponding to, and the structure of the clause for the kana character string to be converted by the kana-kanji converting means. Among comprises a conversion control means for comparing a group likelihoods corresponding to word belongs groups constituting the candidate to determine the structure of the phrase to be translated.

[0015]

According to the present invention, for example, a group of words having a close relationship in terms of meaning is stored as the same group, and a group likelihood indicating the degree of ease of appearance of words belonging to the group is given to each group. Set and update the group likelihood according to the group to which the word used in Kana-Kanji conversion belongs. In Kana-Kanji conversion, the word with the largest group likelihood value corresponding to the group to which the word belongs is converted from the homonyms. The result.

[0016]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described in detail below with reference to the drawings.

FIG. 1 shows an example of the overall configuration of the present invention.

In the configuration shown in the figure, the CPU is a microprocessor, which performs arithmetic operations for character processing, logical judgments, etc., an address bus AB, a control bus CB,
The respective components connected to those buses are controlled via the data bus DB.

The address bus AB is a microprocessor C
An address signal for instructing a component to be controlled by the PU is transferred. The control bus CB transfers and applies a control signal of each constituent element to be controlled by the microprocessor CPU. The data bus DB transfers data between the respective constituent devices.

Next, the ROM is a read-only fixed memory. The PA provided in the ROM is shown in FIGS.
Is a program area in which the procedure of control by the microprocessor CPU described later is stored. Also, NK
Is a microprocessor C described later with reference to FIGS.
This is a program area in which the procedure of controlling the field network by the PU is stored.

The RAM is a writable random access memory having a structure of 1 word and 16 bits, and is used for temporary storage of various data from each constituent element. RAM
Is a word dictionary WDIC, an example dictionary YDIC, a field dictionary B
DIC, field reading dictionary BYDI, field likelihood BYUD, field internal state BNAI, field external input BGAI, field weight BWAI, field likelihood fixed flag BKOT, field mode BMOD, latest use word list WLIS.

The word dictionary WDIC stores word information for performing kana-kanji conversion, and details will be described later with reference to FIG.

The example dictionary YDIC stores information on word pairs that are often used in a semantically related manner in order to improve conversion accuracy of Kana-Kanji conversion, and details will be described later with reference to FIG.

The field dictionary BDIC is a collection and storage of words belonging to a specific field among the words stored in the word dictionary WDIC, and details will be described later with reference to FIG.

The field reading dictionary BYDI collects words belonging to a specific field among words having no homonyms,
It is stored in association with reading, and details will be described later with reference to FIG.

The field likelihood BYUD stores, for each field, information indicating the depth of the relationship between the content of the text currently input by the kana-kanji conversion and each field, and is stored in the field network described later in FIG. Make up a part. Details will be described later with reference to FIGS. 6 and 7.

The field internal state BNAI constitutes a part of the field network described later with reference to FIG. 6, and stores the value of the internal state of each node of the field network.
Details will be described later with reference to FIGS. 6 and 7.

The field external input BGAI constitutes a part of the field network, which will be described later with reference to FIG. 6, and stores the value of the external input to each node of the field network. Details will be described later with reference to FIGS. 6 and 7.

The field weight BWAI stores the weight of a link between fields and constitutes a part of a field network described later in FIG. Details will be described later with reference to FIGS.

The field likelihood fixed flag BKOT stores a flag indicating whether to change the value of the field likelihood BYUD for each field, and constitutes a part of the field network described later in FIG. Details are shown in Figs.
Will be described later.

The field mode BMOD is a flag indicating whether or not to process fields in Kana-Kanji conversion, which will be described later. If this value is 1, it indicates to process fields, and if it is 0, the field processing is performed. Indicates not to do.

The most recently used word list WLIS is an array for storing recently used words, and is stored in order from the most recently used word. As the information to be stored, a word code described later is stored. Each time a word is used as a conversion result in kana-kanji conversion, the word code of that word is stored at the beginning of the list, and the old information is shifted backward. The words stored in this list are treated as short-term learned words as described later.

The KB is a keyboard, which is provided with various function keys such as an alphabet key, a hiragana key, a katakana key and other character symbol input keys, and a cursor movement key for instructing cursor movement. Function keys include KID, SET, JUS,
It has KET, CHG, CLR, and MOD.

KID is a key for instructing the start of kana-kanji conversion. SET is a key for activating various environment setting functions for field processing. JUS is a key that activates the function of creating an address book. KET is a key for determining and instructing a candidate to be converted from among homonymous synonyms for kana-kanji conversion. CHG is a key for designating the field of a sentence input by the operator. CLR is field likelihood BY
This key clears the UD value. The MOD is a key for switching the mode of whether or not to execute the processing of the field.

DISK is an external memory for storing document data and the like. Document data and the like are stored in the external memory DISK as needed, and the stored data is called when necessary by an instruction from the keyboard.

CR is a cursor register. The CPU can read and write the contents of the cursor register. The CRT controller CRTC, which will be described later, displays a cursor at a position on the display device CRT corresponding to the address stored here.

DBUF is a display buffer memory,
Stores the pattern of data to be displayed.

The CRTC plays a role of displaying the contents stored in the cursor register CR and the buffer DBUF on the display CRT.

The CRT is a display device using a cathode ray tube or the like, and the display pattern of the dot configuration and the display of the cursor on the display device CRT are controlled by the CRT controller.

Further, CG is a character generator, which stores patterns of characters and symbols to be displayed on the display device CRT.

The character processing apparatus of the present invention comprising the above components operates in response to various inputs from the keyboard KB, and when an input from the keyboard KB is supplied, an interrupt signal is first sent. Sent to the microprocessor CPU, which microprocessor CP
U reads out various control signals stored in the ROM,
Various controls are performed in accordance with those control signals.

FIG. 2 is a diagram for explaining an example of the structure of the word dictionary WDIC. As information on one word, reading, notation, part of speech, and word ID are stored. The word ID is a unique number assigned so that the word ID can be used to distinguish it from other words. For example, in the dictionary shown in FIG. 2, the reading is “hot”, the notation is “hot”, the part of speech is “adjective”, and the word ID is “123”.
The word "4" is stored. The information of each word is sorted and stored by reading.

FIG. 3 is a diagram for explaining a configuration example of the example dictionary YDIC. Example dictionaries are semantically connected,
It stores two commonly used words together with their connection. For example, in the commonly used sentence "summer is hot", the two words "summer" and "hot" are semantically linked by the particle "ha", and such information is stored in the example dictionary. Is stored. These pieces of information are stored by associating the word IDs of the two words with the connection relationship. For example, in the example dictionary shown in FIG. 3, the word “summer” indicated by the word ID “34567”, the word “hot” indicated by the word ID “1235”, and the connection relation “ha” are associated with each other. It is stored and indicates that "summer" and "hot" are semantically connected by the connection relation "ha".

FIG. 4 is a diagram for explaining a configuration example of the field dictionary BDIC. The field dictionary BDIC stores information on which field and degree the word or phrase is related to. For example, it can be said that the word “trial” is closely related to the field of “law”, but the field dictionary B
The DIC stores such information. In the field dictionary BDIC, information on the relationship between such words and fields is stored in association with the word ID, field and certainty factor.

For example, the field dictionary B shown by FIG.
In the DIC, the word ID "12346", the field "law", and the certainty factor "8" are stored in association with each other, so that the word "trial" indicated by the word ID "12346" corresponds to the field "law". 8 ”indicates that they are related. Here, the certainty factor is a numerical value indicating the degree of the relationship between the word and the field, and the larger the value, the deeper the relationship between the word and the field. It is also possible to take a negative value as the certainty factor. Taking a negative value positively indicates that the word and the field are completely irrelevant, and the smaller the value, the higher the degree of irrelevance. In other words, it indicates that the word is less likely to be used in the field. For example, in the field dictionary BDIC shown in FIG. 4, the word ID “89
The word "yu" represented by "012" is related to the field "politics" with a certainty factor "-2".
A certainty factor of "2" indicates that it is irrelevant.

The field dictionary BDIC can also describe that one word is related to a plurality of fields. For example,
The word "trial" represented by the word ID "23456" is related to the field "law" with a certainty factor "10", and is related to the field "politics" with a certainty factor "2". ..

FIG. 5 is a diagram for explaining an example of the structure of the field reading dictionary BYDI. Like the field dictionary BDIC, the field reading dictionary BYDI stores information on which field and degree the word or phrase is related to. To the field reading dictionary BYDI, homonyms do not exist. Only information about words (phrases) is stored.
Therefore, the word in this dictionary can be specified only by reading, and the field dictionary BDIC uses the word ID as the heading, while the reading is used as the heading. That is, in the field reading dictionary, the reading of a word (phrase), the field, and the certainty factor are stored in association with each other. For example, in the field dictionary shown in FIG. 4, information indicating that the reading "saiban" is related to the field "law" with a certainty factor "10" is stored.

FIG. 6 is a diagram for explaining a configuration example of the field network. In the field network, a node is associated with each field, and each node forms a network structure in which nodes are interconnected via links. In the example shown by FIG. 6, there are “address book”, “law”, “politics”, “cooking”, etc. as fields, and the nodes corresponding to each are mutually connected via a link.

Then, each node corresponding to a field outputs an output value O _i representing the likelihood of the field i (referred to as the field likelihood).
Have each. This output value O _i is basically the field i.
The value is increased when a word belonging to is selected, and the higher the value, the more the text input at that time is related to the field i. The increased output value is updated when the value O _j of another node connected via the link is updated, as will be described later. The output value of this node is updated in a chain. The output value of this node is stored by the field likelihood BYUD. The field likelihood BYUD has a configuration as shown in FIG. 7, for example, and the node 1 corresponding to the field
This is an array in which an area is secured for each one, and the output value of the node is stored in that element.

A link connecting a plurality of nodes to each other has a value (link weight) w _ij indicating the depth of the relationship between the nodes (field i and field j). The link weights are, for example, −1 ≦ w _ij ≦ 1, w _ii = 0, w _ji =
It is given within a range that satisfies the condition of w _ij . Link weight w
_ij takes a larger value as the relationship between the fields i and j is deeper, that is, the words belonging to the field i and the words belonging to the field j are more likely to co-occur. The link weight is stored by the field weight BWAI. Category weight B
The WAI has an area reserved for each weight node of the link, and has a configuration as shown in FIG. 8, for example. Since w _ji = w _ij here, it is represented by a triangular matrix,
It is assumed that the weight of the corresponding link is stored in each element. For example, the element a in FIG. 8 stores the weight of the link between the “politics” field and the “law” field.

The output value O _i of the node corresponding to the field _i is
By calculation of the program stored in the ROM by the microprocessor CPU, for example, O _i (t + 1) = f (u _i (t + 1)) u _i (t + 1) = δu _i (t) + (1 −δ) Σ w _ij O _j (t) + θ _i (t) where O _i (t) is the output value of node field i at time t and u _i (t) is the internal state of node field i at time t. the value f is the upper limit, the [delta] monotonically increasing function with a lower limit, the constant of _{0 ≦ δ ≦ 1 θ i (} t) , when the word belongs to the field i is selected theta _i (t)
It is updated by a process such as an external input value ≧ 0, θ _i (t) ≦ 0 when not selected.

In the above second equation, u _i is the value of the internal state of the node corresponding to the field i, and this value is stored by the field internal state BNAI. Field internal condition BNAI
Has an arrangement as shown in FIG. 7, for example, and is an array in which an area is secured for each node corresponding to a field, and the value u _i of the internal state of the node is stored in that element. To be done. Furthermore, the first term in the second equation is the moment of inertia that suppresses a sudden output change of the output value O _i of the node. The second term in the second equation is a term indicating the influence of the output value of the adjacent node on the output value of the own node, and the link weight w _ij is the field weight BWAI as described above.
Stored by. Further, O _j is stored by the field likelihood BYUD. Next, the third term θ _i is to increase the output value of the node of the corresponding field when the word belonging to the field i is selected, and to gradually decrease the output value of the node when it is not selected. Of the field external input BGAI. The field external input BGAI has a configuration as shown in FIG. 7, for example, and is an array in which an area is secured for each node corresponding to the field, and the value of the external input of the node is set in that element. Is stored. A method of obtaining θ _i will be described later.

Further, the monotonically increasing function f in the above equation 1
Is for keeping the output value O _i of the node within a specific range. For example, the sigmoid function f (x) = 1 / (1 + e
xp (-x)). In the case of such a function, in order to obtain the function value, a function value table in which the variable value and the function value are associated with each other may be prepared in the memory.

Therefore, the above program is composed of the field internal state BNAI, the field weight BWAI, the field likelihood BYUD,
Internal state u at time t stored in field external input BGAI
_i, the weight w _ij of the link, the output value O _j of each node, in accordance with the second equation from the external input value theta _i, determine the time t + 1 of u _i, refers to the function value table of the function f the value of the u _i Thus, the output value O _i of the node at time t + 1 is obtained.

Here, each node is provided with a flag as to whether or not to change the output value, and for the node whose flag value is 1, the output value of the node is updated by the above equation. No processing is done. Only when the value of this node is 0, the output value of the node is updated. This flag is stored by the field likelihood fixed flag BKOT. The field likelihood fixed flag BKOT has, for example, a configuration as shown in FIG. 7, is an array in which an area is secured for each node corresponding to the field, and the value of the flag is stored in that element. To be done. Normally, the value of this flag is 0, but when the operator specifies a field by changing the field setting described later, the value of this flag is set to 1 for that field. This is used to keep the likelihood of the field designated by the operator (that is, the output value of the node) constant.

The microprocessor CPU sequentially executes such update processing, and updates the values of the nodes in the field network in a chained manner in accordance with the link structure. The updating process of the node value is sequentially and sequentially executed independently of the Kana-Kanji conversion process controlled and executed by the CPU according to the program in the program area PA. However, the third term θ _i (t) of the second expression is determined according to the selection information each time a word is selected in the kana-kanji conversion processing, as described later. That is, it is updated by the kana-kanji conversion processing. Then, the kana-kanji conversion processing proceeds according to the state of the field network at that time, that is, the field likelihood BYUD. The procedure for updating the output value of each node of the field network as described above is stored in the network management procedure NK.

FIG. 9 shows an example of a field designation screen in which the operator inputs a field by inputting the CHG key after the SET key. As shown in the figure, for each field, a "number", a "field name", and a "field designation" indicating the status (designated / not designated) of the field designation are displayed in association with each other and designated by the operator. It shows the situation in the field. For example, in FIG. 9, the fields are designated in the fields of "politics" and "law", and the fields are not designated in the fields of "address book", "sports", and "cooking". This means that the operator is trying to input a sentence in the "politics" or "law" field.

A procedure for executing the above operation according to the program stored in the program area PA will be described with reference to a flowchart.

FIG. 10 is a flow chart showing the operation of the character processing apparatus of the present invention.

Step S10-1 is a process for making various initial settings of the character processing device of the present invention, and such initial setting is a process generally performed in a character processing device of the same type. In this embodiment, in particular, the field likelihood BYUD,
The value of each element of the field internal state BNAI, field external input BGAI, and field likelihood fixed flag BKOT is initialized to 0, or the field mode BMOD value is initialized to 0. Further, the process according to the update processing procedure of the node of the field network stored in NK is operated in parallel with this process. When the processing is completed, step S10
-2.

Step S10-2 is a process for fetching data from the keyboard. The type of the key fetched in step S10-3 is determined, and the process branches to the processing routine for each key.

When the KID key is input, step S1
Branching to 0-4, the kana-kanji conversion processing described in detail with reference to FIG. 11 is performed. After the conversion process is performed, step S
Proceed to 10-8.

When the SET key is pressed, step S
The process branches to 10-5, and the field environment setting process described in detail with reference to FIG. 24 is performed. Then, it progresses to step S10-8.

When the JUS key is pressed, step S
Branching to 10-6, the address book creation process is performed. In the address book creation process, first, the value of the element corresponding to the address book field of the field likelihood fixed flag BKOT is set to 1, and the values of the other elements are set to 0. That is, the output value of the node corresponding to the address book field in the field network is set so as not to change.

Next, the internal state value and the output value of the node corresponding to the address book field in the field network are set to the maximum possible output values, and the internal state value and the output value of the other nodes are set. Set to 0. That is, in the field likelihood BYUD and the field internal state BNAI, the value of the element corresponding to the address book field is set to the maximum value,
Set the values of the other elements to 0. Further, the values of all elements of the field external input BGAI are set to 0, and the values of external input to each node in the field network are set to 0.

Then, the address book preparation process is performed.
The address book creating process is a process that is generally performed in the same type of character processing device and is well known, so a detailed description thereof will not be given. However, in the address book creation process, the same process as the Kana-Kanji conversion performed in step S10-4 is performed, but since the field likelihood and the field likelihood fixed flag value are set as described above. , Network management procedure N
According to K, even if the microprocessor CPU operates the field network independently of the kana-kanji conversion processing, the likelihood of the address book field always maintains the maximum value. It will be converted in preference to words belonging to the field. When the address book creation process is completed, the process proceeds to step S10-8.

For other keys, step S10-7.
And other processing such as insertion and deletion that is performed in a normal character processing device is performed. These processes are processes that are generally performed in the same type of character processing device, and are known, and therefore will not be described here. After the processing is performed, the process proceeds to step S10-8.

Step S10-8 is a display process for displaying the changed portion as a result of the above-mentioned processes. This is a generally-used process of expanding one character of data in a document into a character pattern each time it is read and outputting it to a display buffer. When the processing is completed, step S1
Return to 0-2.

FIG. 11 is a detailed flowchart of "Kana-Kanji conversion" in step S10-4.

Step S11-1 is a process for fetching data from the keyboard. It is determined whether or not the type of the key fetched in step S11-2 is read, and if read, the process branches to step S11-3, and if not read, the process branches to step S11-4.

In step S11-3, the input readings are stored in the buffer in the order of input. This process is a process generally performed in the same type of kana-kanji conversion process. When the storage of the input reading is completed, the process proceeds to step S11-1.

In step S11-4, step S11
In -3, the reading stored in the buffer is converted into Kanji. This process will be described in detail with reference to FIG. After that, it advances to step S11-5.

In step S11-5, step S
Correct the Kanji converted in 11-4. This process is shown in Figure 1.
5 will be described in detail. When the processing is completed, it returns.

FIG. 12 is a detailed flowchart of the "reading conversion" in step S11-4.

Step S12-1 is a process for extracting all the combinations having the best reading length from the combinations of two consecutive phrases stored in the buffer from the reading conversion target position. This processing is a method called the two-segment longest-match method, and the word dictionary WD
This is a process to extract grammatically correct phrases by searching ICs, and to extract a combination with the longest reading length of the two phrases from the combination of two consecutive phrases, which is a kana-kanji conversion process of the same kind. Since this is a process generally performed in, the detailed description will not be given.

In this process, there are cases where two consecutive phrases cannot be extracted and only one phrase can be extracted. In this case, only the longest clause is extracted. Step S12-2 is a process for checking this. If only one phrase can be extracted, all the readings stored in the buffer have been extracted as a phrase, so step S12-2.
Branch to 12-6. If two consecutive phrases have been extracted, the process branches to step S12-3.

Step S12-3 is step S12-
In the combination of the two clauses extracted in 1, whether there is a difference in the delimiter position of the two clauses, that is,
Check if there are multiple candidates for delimiter positions. If there is more than one, branch to step S12-4,
If there is only one delimiter position, the process branches to step S12-5.

In step S12-4, only one of the most appropriate delimiter positions is selected from a plurality of different delimiter positions and the determined position is determined. After the determination, the determined delimiter position is different. The combination of two clauses separated by the position is excluded from the conversion candidates. This process is shown in FIG.
Will be described in detail. When the processing ends, step S12-5
Go to.

In step S12-5, the conversion result is determined for the first phrase of the two consecutive phrases, the determined conversion result is output, and the reading of the conversion target is moved to the head position of the succeeding phrase. Processing. This process will be described in detail with reference to FIG. When the process is completed, the process proceeds to step S12-1.

Step S12-6 is step S12-
The conversion result is determined for the phrase extracted in 1.
This is a process of outputting the conversion result. Since these processes are the same as the processes performed in step S12-5,
No detailed description will be given. When the processing is completed, it returns.

FIG. 13 is a detailed flow chart of the process of "delimitation determination" in step S12-4.

Step S13-1 is a process of narrowing down a plurality of delimiter positions according to an example. That is, if a combination of two independent words forming two consecutive phrases is stored in the example dictionary YDIC and the connection relation described in the example dictionary is satisfied, the delimiter between the two phrases is an example. There is a break.

Further, a combination of the independent word which has already been converted and output and which constitutes one of the two consecutive bunsetsus is stored in the example dictionary YDIC, and the combination of the independent words is also stored in the example dictionary. If the described connection relation is satisfied, the break between the two clauses is regarded as a break with an example. In this way, it is checked whether or not there is an example for each delimiter, and if there is a delimiter with an example, a combination of two clauses having a delimiter without an example is excluded from the candidates.

In this way, it is possible to narrow down the delimiter position depending on the example. For example, for the reading "Hitoto Hanasu", two clauses "Hitoto (Human) / Hanasu (Speak)" and "Hitoto (Human) / Eggplant (Eggplant)" are possible. However, when the example in which "person" and "speak" are combined in the combination relation "to" is stored in the example dictionary YDIC, and the example in which "person" and "eggplant" are combined as described above is not in the dictionary , There is an example in the section of "Hitoto (with person) / Hanasu (speak)", but there is no example in the section of "Hitoto (with person) / Eggplant (eggplant)". As a result of the processing, only the delimiter of "people (with people) / release (speaking)" is left. When this process ends, the process proceeds to step S13-2.

The step S13-2 is the step S13-
In step 1, it is a process of determining whether or not the break position can be determined to be one according to the example. If there are a plurality of break candidates, the process branches to step S13-3, and if there is only one break candidate. If you return.

Step S13-3 is a process of narrowing down a plurality of delimiter positions depending on the field. This is a process of determining the likelihood of two phrases based on the state of the category likelihood, that is, the relationship between the state of the category network and the category to which the words forming the two phrases belong, and narrowing the delimiter based on the certainty. This process will be described in detail with reference to FIG. When this process ends, the process proceeds to step S13-4.

The step S13-4 is the step S13-
3 is a process for determining whether or not one break position can be determined depending on the field. If there are a plurality of break candidates, the process branches to step S13-5, and if there is only one break candidate. If you return.

Step S13-5 is a process of narrowing down a plurality of delimiter positions by short-term learning. That is, if either one of the two independent words forming the two phrases is stored in the most recently used word list WLIS, the segment of the two phrases is a segment with short-term learning. In this way, it is checked whether or not there is short-term learning for each segment, and if there is a segment with short-term learning, a combination of two clauses having a segment without short-term learning is excluded from candidates.

For example, in contrast to the reading "here is a kimono", 2 is "here / has a kimono (footwear)".
It is possible to have two clauses, a clause and "Here / Kimono (Kimono)". At this time, the latest word list WLIS
If "Kimono (Kimono)" is registered and "Here" and "Footwear" are not registered, there is a short-term learning at the "Here / Kimono (Kimono)" section.
There is no short-term learning in the "here / footwear (footwear)" section, and as a result of this processing, only the "here / kimono (kimono) section" remains. When this process ends, the process proceeds to step S13-6.

Step S13-6 is step S13-
In step 5, it is a process of determining whether or not one break position can be determined by short-term learning. If there are a plurality of break candidates, the process branches to step S13-7, and only one break candidate exists. If there is, return.

Step S13-7 is a process for determining a break by a method other than the above, for example, a process for determining a break according to the frequency of use of words. Such a process is a process generally performed in the same type of kana-kanji conversion process and is well known, so a detailed description thereof will not be given. When the processing is completed, the number of divisions is determined to be one, so the process returns.

FIG. 14 is a detailed flowchart of the "determine by field" process in step S13-3.

First, in step S14-1, the process branches according to the value of the field mode BMOD. Field mode BM
If the value of OD is 1, it is the mode in which the field processing is performed, and the process branches to step S14-2. Field mode BMO
If the value of D is 0, it is a mode in which no field processing is performed, and therefore the processing returns without performing the subsequent processing.

At step S14-2, the two bunsetsu targeted for the processing at steps S14-4 to S6 are taken out one by one from the two bunsetsu candidates. Step S
In 14-3, if unprocessed candidates can be extracted in step S14-2, the process branches to step S14-4, and if all candidates have been extracted, the process branches to step S14-3 and step S14- Proceed to 7.

In step S14-4, for the first bunsetsu among the two bunsetsu, the bunsetsu likelihood by the field, that is, the degree of certainty of the bunsetsu by the field processing is obtained. Details of this processing will be described in detail with reference to FIG. When the processing is completed, the process proceeds to step S14-5.

In step S14-5, of the two clauses,
For the bunsetsu behind, find the bunsetsu likelihood by field. Since this processing is also performed in the same manner as step S14-4, the details will be described with reference to FIG. When the processing is completed, the process proceeds to step S14-6.

In step S14-6, step S
14-4, the bunsetsu likelihood by the field for the first bunsetsu and the bunsetsu likelihood by the field for the succeeding bunsetsu, which are found in step S14-5, are added. This value is
It is the likelihood depending on the field of the target two clauses. Then, this value is stored and saved in association with the target two clauses. When the processing ends, step S14-2
Go to.

In step S14-7, step S
The candidates for the two phrases are narrowed down according to the likelihood of the two phrases stored and saved in 14-6. For example, 2
The other two clauses are excluded from the candidates, leaving only the candidate having the highest likelihood value depending on the clause field. When the processing is completed, it returns.

FIG. 15 is a detailed flowchart of the "field clause likelihood (1)" processing of step S14-4 and the "field clause likelihood (2)" processing of step S14-5.

Step S15-1 is a process for searching whether or not the independent word forming the target clause is registered in the field dictionary BDIC. Field dictionary BD
The IC is sequentially searched from the beginning, and each time one piece of information on the registered field is found, the process branches to step S15-3 in step S15-2. Field dictionary BDIC
When the search is completed to the end of, the process branches to step S15-5 in step S15-2.

In step S15-3, step S
The likelihood of the clause according to the field is obtained from the information of the field searched in 15-1 and the output value of the field network, that is, the field likelihood. Assuming that the information of the detected field is the field i and the certainty factor is c _i , the likelihood BY _i of the target bunsetsu by the field i is the output value () of the node corresponding to the field i of the field network. Ie field likelihood) O
_{Using i} , for example, the following formula is used.

BY _i = O _i c _i (for c _i ≧ 0) BY _i = 0 (for c _i <0) In this way, when the likelihood BY _i of the phrase by the field i is obtained, The larger the positive value and the larger the value (that is, the more the input sentence is related to the field i), and the larger the certainty factor that the independent words constituting the clause belong to the field i, the larger ( That is, the deeper the independent word is related to the field i), the larger the value of BY _i becomes. When the likelihood of the phrase according to the field is obtained, the process proceeds to step S15-4.

In step S15-4, step S15
-3 is a process of obtaining the maximum value of the likelihood of a phrase according to the field obtained. In this process, the old maximum value is compared with the value obtained in step S15-3, and if the value obtained in step S15-3 is larger, that value is saved as a new maximum value. When the processing is completed, the process proceeds to step S15-1.

In step S15-5, the corresponding field is obtained by the part of speech stored in the word dictionary, and the likelihood of the phrase by the field is obtained. That is, the part of speech of the independent word of the target phrase is obtained from the word dictionary, and if the part of speech is associated with a specific field, the word is treated as belonging to that field. For example, if the word "Ikeda" is read and the part of speech of the word "Ikeda" is "person's name" and the part of speech "person's name" is associated with the field "Address Book", the word "Ikeda" is It belongs to the "address book".

In this case, the certainty factor has a constant value, for example. Then, according to the calculation formula used in step S15-3, the likelihood by clause is obtained. If the part of speech of the target word is not associated with a specific field, the likelihood of the bunsetsu depending on the field is 0. When the likelihood of the bunsetsu depending on the field is obtained from the part of speech as described above, step S
Proceed to 15-6.

Step S15-6 is step S15-
The maximum value of the likelihood of the phrase by the field obtained in 4 is compared with the likelihood of the phrase by the field obtained in step S15-5. If the value obtained in step S15-5 is larger, the maximum value is set to that value. Is a process of updating to. When the processing is completed, it returns.

FIG. 16 shows steps S12-5 and
It is the flowchart which detailed the "conversion result determination" process of step S12-6.

In step S16-1, the conversion result of the target clause is determined. This process will be described in detail with reference to FIG. When the processing is completed, the process proceeds to step S16-2.

The step S16-2 is the step S16-
This is a process of outputting the conversion result in accordance with the clause structure determined in 1. This process is a process generally performed in the same type of kana-kanji conversion process, and a detailed description thereof will not be given.

In step S16-3, the process branches depending on the value of the field mode BMOD. If the value of the field mode BMOD is 1, it is the mode in which the field processing is performed, so the process branches to step S16-4, and the value of the field mode BMOD is 0.
If so, the mode is a mode in which field processing is not performed, and the process returns.

In step S16-4, the value of the external input to the field network is updated as necessary by reading the clause. This process will be described in detail with reference to FIG. When the processing is completed, it returns.

FIG. 17 is a detailed flowchart of the "conversion word determination" process of step S16-1.

Step S17-1 is a process of narrowing down the words to be converted according to the example. That is, if a combination of a target phrase and two independent words that form another phrase is stored in the example dictionary YDIC, and if the combination relation described in the example dictionary is satisfied, the target phrase is Consider it as a phrase with an example. In this way, it is checked whether or not there is an example for a bunsetsu candidate, and if there is a bunsetsu with an example, a bunsetsu with no example is removed from the candidates.
In this way, it is possible to narrow down the candidate bunsetsu to be converted according to the example. This process is a process generally performed in the same type of kana-kanji conversion process. When this process ends, the process proceeds to step S17-2.

The step S17-2 is the step S17-
1 is a process of determining whether or not one bunsetsu candidate to be converted can be determined. If there are a plurality of bunsetsu candidates, the process branches to step S17-3, and only one bunsetsu candidate is present. If so, return.

Step S17-3 is a process of narrowing down a plurality of clause candidates by field. Depending on the state of the field likelihood, that is, the state of the field network and the field to which the words that make up the target clause belong,
This is a process of obtaining the certainty of a phrase and narrowing down the candidates of the phrase based on the certainty. This process will be described in detail with reference to FIG. When this process ends, step S17-
Go to 4.

The step S17-4 is the step S17-
3 is a process of determining whether or not one bunsetsu candidate to be converted can be determined. If there are a plurality of bunsetsu candidates, the process branches to step S17-5, and only one bunsetsu candidate exists. If there is, return.

Step S17-5 is a process of narrowing down a plurality of clause candidates by short-term learning. That is, the independent words that make up the phrase are the most recently used word list WLI
If it is stored in S, it is assumed that the phrase has been short-term learned. In this manner, it is checked whether or not each phrase candidate has been short-term learned, and if there is a phrase that has been short-term learned, the phrase that has not been short-term learned is removed from the candidates. Such processing is generally performed in the same type of kana-kanji conversion processing. When this process ends, the flow proceeds to step S17-6.

The step S17-6 is the step S17-
5 is a process for determining whether or not one bunsetsu candidate to be converted can be determined. If there are a plurality of bunsetsu candidates, the process branches to step S17-7, and only one bunsetsu candidate is present. If so, return.

Step S17-7 is a process for determining the structure of the phrase to be converted by a method other than the above, and is a process for determining the word to be converted according to the frequency of words, for example. Such a process is a process generally performed in the same type of kana-kanji conversion process and is well known, so a detailed description thereof will not be given. When the processing is completed, the structure of the bunsetsu to be converted is determined to be one, so the process returns.

FIG. 18 is a detailed flowchart of the "determine by field" process in step S17-3.

In step S18-1, the process branches depending on the value of the field mode BMOD. If the value of the field mode BMOD is 1, it is the mode in which the field processing is performed, so the process branches to step S18-2, and the value of the field mode BMOD is 0.
If so, the mode is a mode in which field processing is not performed, and the process returns.

At step S18-2, the target bunsetsu is taken out one by one from the bunsetsu candidates. If the clause can be extracted in step S18-2, step S18-4 is executed in step S18-3.
Branch to, complete extraction of all candidates, step S1
If a new phrase cannot be extracted in 8-2, the process branches to step S18-6 in step S18-3.

In step S18-4, for the target bunsetsu, the bunsetsu likelihood by the field, that is, the degree of certainty of the bunsetsu by the field processing is obtained. This processing is the same as the processing in steps S14-4 and S14-5, and since it has already been described in detail with reference to FIG. 15, it will not be particularly described here. When this process ends, the flow proceeds to step S18-5.

In step S18-5, step S
The phrase likelihood by the field obtained in 18-4 is stored and saved in association with the target phrase. When this process ends, the process proceeds to step S18-2.

In step S18-6, bunsetsu candidates are narrowed down based on the likelihood of the bunsetsu field stored and saved in step S18-5. For example, only the candidate having the highest likelihood value depending on the field of the phrase is left and other phrases are excluded from the candidates. When the processing is completed, it returns.

FIG. 19 is a flow chart for explaining the "reading and updating field" processing in step S16-4.

The step S19-1 is the step S16-
According to the structure of the phrase to be converted determined in 1, the field reading dictionary BYDI is obtained by reading the independent word in this phrase.
Is a process of searching for. From the head of the field reading dictionary BYDI, dictionary items that match the reading of the independent word are sequentially found one by one, and each time it is found, the process branches to step S19-3 in step S19-2. Therefore, when the reading of the independent word in the phrase corresponds to a plurality of items in the field reading dictionary BYDI, the process of step S19-3 is executed for each item. When the field reading dictionary BYDI has been searched to the end, the process returns.

The step S19-3 is the step S19-
This is a process of updating the external input of the field network according to the dictionary item of the field reading dictionary BYDI found in 1. If the content of the dictionary item of the field reading dictionary BYDI found in step S19-1 is field i,
If the certainty factor is c _i , the value of the external input θ _i for the node corresponding to the field i in the field network is expressed by the following equation.

Θ _i = Z _i c _i Here, Z _i is a positive constant set for each field. By updating the external input in this way, c _i
In There positive value, the external input theta _i to node i larger increases, the output value of the node i, i.e., the likelihood O _i of the field i is increased by words belonging to the field i has priority It is easy to convert. Also, c _i is a negative value,
External input theta _i to smaller the node i is smaller becomes small (negative absolute value is large), the output value of the node i, i.e., the likelihood O _i of the field i is reduced, the field i
Words belonging to are hard to convert.

FIG. 20 is a flow chart for explaining the "conversion result correction" processing of step S11-5.

The step S20-1, the step S11-
In this processing, the clauses are taken out one by one from the beginning of the conversion result converted and output by the "reading conversion" in 4, and the operator is inquired whether the conversion result is correct. Each time one clause is taken out, the process branches to step S20-3 in step S20-2. When all the clauses have been taken out, the process returns in step S20-2.

At step S20-3, the data from the keyboard KB is fetched. The entered key is KET
If it is a key, step S20-4
Branch to 20-7. If any key other than the KET key is pressed, the step S20 is performed in the step S20-4.
Branch to -5.

In step S20-5, step S20-
This is a process for obtaining and displaying candidates for homonyms for the phrase extracted in 1. This process will be described in detail with reference to FIG. When the process is completed, the process proceeds to step S20-6.

In step S20-6, step S20-
This is a process in which the operator selects a candidate to be converted from the candidates displayed in 5. For example, step S20
In -5, if each homonym is displayed with a number, the operator can select the candidate to be converted by designating the number and inputting the corresponding numeric key. When a candidate to be converted is selected, step S20
Go to -7.

Step S20-7 is a process of updating the external input to all the nodes of the field network. This process will be described in detail with reference to FIG. When the process is completed, the process proceeds to step S20-8.

Step S20-8 is a process of determining and outputting the conversion result selected by the operator for the target clause. This process is a process generally used in the same type of kana-kanji conversion process and is well known, so a detailed description thereof will not be given. When the processing is over,
It proceeds to step S20-1.

FIG. 21 is a flow chart for explaining the "homogeneous synonym display" processing of step S20-5.

Step S21-1 is a process for extracting all conversion candidates for the target clause. This process is generally used in the same type of kana-kanji conversion process and is well known, so a detailed description thereof will not be given. When the processing is completed, the process proceeds to step S21-2.

The step S21-2 is the step S21-
This is a process of determining the order in which the candidates extracted in 1 are displayed. This process will be described later in FIG. When the process is completed, the process proceeds to step S21-3.

The step S21-3 is the step S21-
This is a process of displaying the conversion candidates extracted in 1 in the order determined in step S21-2. For example, a number is displayed corresponding to each candidate. This process is generally used in the same type of kana-kanji conversion process and is well known, so a detailed description thereof will not be given. When the processing is completed, it returns.

FIG. 22 is a flow chart for explaining the "display order change" processing of step S21-2.

The step S22-1 is the step S21-
This is a process of extracting a candidate having an example from the candidates extracted in 1 and bringing it to the head of the priority order. The determination as to whether or not the example has a certain phrase has been described in step S17-1 and will not be particularly described here. When the processing ends, the process proceeds to step S22-2.

In step S22-2, the field mode BM
Branch depending on the value of OD. If the value of the field mode BMOD is 1, it is the mode for performing the field processing, and thus the process branches to step S22-3. If the value of the field mode BMOD is 0, it is the mode for not performing the field processing, so that the step S22-3 is performed.
It branches to 22-4.

The step S22-3 is the step S22-
From the candidates that are not extracted in step 1, the priority candidate is extracted according to the phrase likelihood depending on the field, and step S2
This is the process of arranging after the candidate that has been prioritized in 2-1. For example, for each target bunsetsu candidate, a bunsetsu likelihood by field is obtained, only candidates whose value is a certain value or more are taken out, and in the descending order of the value, the candidate prioritized in step S22-1 is searched. Line up. The method of obtaining the phrase likelihood according to the field has already been described in detail with reference to FIG. 15, and will not be described here. When the process is completed, the process proceeds to step S22-4.

In step S22-4, step S22-
1. This is a process of extracting a priority candidate by word learning from the candidates not extracted in step S22-3 and arranging it after the priority candidate in step S22-3. For example, if an independent word is stored in the most recently used word list WLIS among the target bunsetsu candidates, the independent word has been subjected to word learning. Place them behind the candidate that was prioritized in -3. When the processing ends, the process proceeds to step S22-5.

In step S22-5, step S22-
This is a process in which the output order is determined by another method for the candidates that have not been extracted in step S22-3 and step S22-4, and the candidates are arranged after the priority candidate in step S22-4. This process is a process generally performed in the same type of kana-kanji conversion process. When the processing is completed, it returns.

FIG. 23 is a flow chart for explaining the "external input update" processing of step S20-7.

In step S23-1, the field mode BM
Branch depending on the value of OD. If the value of the field mode BMOD is 1, it is the mode for performing the field processing, so the process branches to step S23-2. If the value of the field mode BMOD is 0, the mode is not for the field processing, and the process returns.

In step S23-2, the value of the external input to each node of the field network, that is, the value of each element of the field external input BGAI is updated. For example,
If the value of the external input is a positive value, it is updated to a value obtained by subtracting 1 from the value, if it is a negative value, it is updated to a value obtained by adding 1 to the value, and if the value is 0, Do not update. This means that the value of the external input is updated so as to approach 0. When the processing is completed, the process proceeds to step S23-3.

The step S23-3 is a target word,
That is, it is a process of searching the field dictionary BDIC by the independent words that form the clause selected by the operator. A search is performed by word ID from the beginning of the field dictionary BDIC, and whenever dictionary information with a matching word ID is found,
In step S23-4, the process branches to step S23-5. When the search is completed up to the end of the field dictionary BDIC, the process branches to step S23-6 in step S23-4.

The step S23-5 is the step S23-
This is a process of updating the value of the external input to the node of the corresponding field, that is, the field external input BGAI according to the content of the dictionary information found in 3. That is, assuming that the field of dictionary information is field i and the certainty factor is c _i , BG
Set the following value to the value θ _i of the external input of the field i in AI.

Θ _i = X _i c _i Here, X _i is a positive constant set for each field. By updating the external input in this way, c _i
Is a positive value, and the larger it is, the larger the external input to the node i becomes, and the output value of the node i, that is, the field i.
The likelihood O _{i of} is large, and words belonging to the field i are prioritized and easily converted. Also, as c _i is a negative value, the smaller it is, the smaller the external input to the node i is,
The output value of the node i, that is, the likelihood O _i of the field i is small, and words belonging to the field i are hard to be converted. When the processing ends, the process proceeds to step S23-3.

Step S23-6 is a process of updating the value of the external input, that is, the field external input BGAI, by the part of speech of the target word, that is, the part of speech of the independent word forming the clause selected by the operator. As described in step S15-5, the specific part of speech is associated with the specific field, and the certainty factor in that case can have a constant value. In this process, this input is used to update the external input. That is, if the field corresponding to the part of speech is field i and the certainty factor is c _i ,
Using the formula used in step S23-5, the value of the external input value θ _i of the field i is set. When the processing is completed, it returns.

FIG. 24 is a flow chart for explaining the "field environment setting" processing of step S10-5.

Step S24-1 is a process for fetching data from the keyboard. The type of the key fetched in step S24-2 is determined, and the process branches to each key processing routine.

When the CLR key is pressed, step S2
By branching to 4-3 and setting the output value of each node of the field network, that is, the value of each element of the field likelihood BYUD to 0, the field likelihood is cleared. When the processing is completed, it returns.

When the MOD key is pressed, step S2
It branches to 4-4 and switches the value of the field mode BMOD. That is, 0 if the field mode BMOD value is 1.
If it is 0, it is set to 1 to switch whether or not to process the field. When the processing is completed, it returns.

When the CHG key is pressed, step S2
The flow branches to 4-5 to display the designated state of the field at that time, as shown in FIG. The "presence" and "absence" of the field designation are the results of the judgment based on the value of each element of the field likelihood fixed flag BKOT. Here, if the value of the element corresponding to the target field is 1, it is determined to be “present”, and if it is 0, it is determined to be “none”. When the display is finished,
It proceeds to step S24-6.

In step S24-6, the data from the keyboard is further fetched. Then, in step S24-7, it is determined whether or not the taken-in key is a numeric key. If it is a numeric key, the process branches to step S24-8, and if it is not a numeric key, the process branches to step S24-6.

The step S24-8 is the step S24-
This is a process of changing the designation of the field corresponding to the number of the number key taken in in 6. As shown in FIG. 9, if the designation of the field corresponding to the numeral is “present”, it is changed to “none”, and if “no”, it is changed to “present”. Setting the field to "Yes" sets the value of the element of the field likelihood fixed flag BKOT corresponding to the target field to 1, and further sets the value of the element of the field likelihood BYUD corresponding to the target field. Is set to the maximum value. Setting the field designation to "none" sets the value of the element of the field likelihood fixed flag BKOT corresponding to the target field to 0, and further sets the value of the element of the field likelihood BYUD corresponding to the target field. To set the value to 0. When the processing is completed, it returns.

According to this embodiment, even for a word that appears for the first time or a word that does not have a semantic connection, an appropriate word can be effectively used as a conversion result according to the information of the already used word. You can Moreover, the memory capacity and the amount of calculation required for that are small.

That is, in contrast to the above-mentioned word network proposed in Japanese Patent Laid-Open No. Hei 3-22167, one likelihood representing the degree of appearance of a word is provided for each word. In this case, since words are grouped and one likelihood representing the degree of appearance of a word belonging to the group is provided for one group, the likelihood and the value of the link used to update the likelihood, etc. The memory required for is greatly reduced. Also, since the number of likelihoods to be updated is small, the amount of calculation is greatly reduced, and practical kana-kanji conversion can be realized.

[Other Embodiments] The present invention is not limited to the above embodiments.

For example, in the above-mentioned embodiment, the unit of a group of words having a close semantic relation is used as a field, but a unit of a smaller unit may be used. For example, a group of words that are semantically identical, such as "car" and "car," may be used as a unit, or a group of words that have the same semantic attribute, such as "right" and "left", may be used as a unit. Good.

As a means for grouping words,
Although a field dictionary BDIC different from the word dictionary WDIC is used, group information may be stored in the word dictionary WDIC.

The field network may be constructed according to the specifications of the Japanese language processing system.
The setting of the output value of the node, the setting of the value of the link weight, the update processing procedure of the output value of the node, and the like may be determined according to the specifications.

Further, although the output value update processing of the node of the field network is controlled by the CPU which also executes the control based on the procedure stored in the PA, it is dedicated to the output value update processing. Hardware may be used.

Although the certainty factor is used as a measure of the degree of relationship between a word and a field, the certainty factor is treated as having a positive constant value in order to simplify the processing, and the certainty factor is omitted. May be.

Further, when deciding a paragraph break in kana-kanji conversion, as shown in FIG. 13, the decision by example, the decision by field, the decision by short-term learning and the decision by other methods are applied in this order. I have decided,
This decision algorithm can be variously modified. For example,
The determination may be performed by applying in the reverse order of this order, or each determination method may be performed for all candidates and a predetermined weighting may be given to each processing result to make a comprehensive determination.

Further, one of the methods for determining the segment breaks in the kana-kanji conversion is the determination by short-term learning. In the above embodiment, if either one of the two phrases has been short-term learned, it is prioritized. However, in this method, a candidate whose both clauses have been short-term learned and a candidate whose only clause has been short-term learned are treated equally. Therefore, if there is a candidate in which both clauses are short-term learned and a candidate in which only one clause is short-term learned, only the candidate in which both clauses are short-term learned may be prioritized. .. For example, in response to the reading "here, kimono", two clauses, "here / footwear (footwear)" and "here / kimono (kimono)" are possible. At this time, if “here” and “kimono (kimono)” are registered in the latest word list WLIS, and “footwear” is not registered, “here / kimono (kimono)” delimiter Both bunsetsu are learned for a short period of time, but only one bunsetsu is learned for a short time in the delimiter of "here / footwear (footwear)". Only the break of "))" will be left.

Further, when determining a word to be converted in the kana-kanji conversion, as shown in FIG. 17, the determination by the example, the determination by the field, the determination by the short-term learning, and the determination by other methods are applied in this order. I have decided,
This determination algorithm can also be modified in various ways, as in the case of delimitation determination. For example, it may be applied in the reverse order of this order to make a decision, or each decision method may be applied to all candidates and a predetermined weight may be given to each processing result to make a comprehensive decision. Good.

Further, in the kana-kanji conversion, when determining the display order of all the homonymous conversion candidates for a phrase, as shown in FIG. 22, determination by example, determination by field, determination by short-term learning and other The determination by the method is applied in this order to determine, but this determination algorithm can be variously modified. For example, it may be applied in the reverse order of this order to make a decision, or each decision method may be applied to all candidates and a predetermined weight may be given to each processing result to make a comprehensive decision. Good.

Further, in the above description, the updating process of the value of the node of the field network is executed independently of the kana-kanji conversion, but it may be executed in synchronization with the specific process of the kana-kanji conversion. For example, step S2
After updating the value of the external input to the node of the field network at 0-7, the updating process of the value of the node may be sequentially executed in a chain.

Further, in the above description, the nodes of the field network are mutually connected, but it is not always necessary that all nodes are mutually connected. Further, the updating of the value of the node does not have to be performed in a chain. For example, when the word belonging to the field i with the certainty factor c _i is determined as the word to be converted, only when the external input is updated in step S20-7, the node expression The value may be updated.

O _j (t + 1) = f (u _j (t + 1)) u _j (t + 1) = δu _j (t) + (1-δ) V _ji c _i where O _j ( t) is an output value u _j (t) of the node field j at time t, an internal state value f of the node field j at time t, a monotonically increasing function δ having upper and lower limits, and 0 ≦ δ ≦ 1 Here, V _ji is the degree of affecting the value of the node corresponding to the field j when the word of the field i is determined as the word to be converted, and the deeper the relationship between the field i and the field j is, That is, the larger the co-occurrence of the word belonging to the field i and the word belonging to the field j, the larger the value.

In addition, the present invention can be variously modified and implemented without departing from the scope of the invention.

The present invention is applicable not only to a single device but also to a system composed of a plurality of devices.
It goes without saying that it can be realized by providing software to the device or system.

[0178]

As is apparent from the above description, according to the present invention, a group likelihood that represents a degree of easiness of appearance of words belonging to a group by grouping a group of words having a semantically close relationship. Is associated with each group, and the group likelihood is updated according to the group to which the used word belongs, and in the Kana-Kanji conversion, the word with the largest group likelihood value corresponding to the group to which the word belongs among the homonyms is converted. So, the reading that first appeared,
Even for the reading of words having no semantic connection, it is possible to determine an appropriate word as a conversion result based on the information of already used words with a small memory capacity and a small calculation amount.

[Brief description of drawings]

FIG. 1 is a block diagram of the overall configuration of the present invention.

FIG. 2 is a diagram showing a configuration example of a word dictionary WDIC.

FIG. 3 is a diagram showing a configuration example of an example dictionary YDIC.

FIG. 4 is a diagram showing a configuration example of a field dictionary BDIC.

FIG. 5 is a diagram showing a configuration example of a field reading dictionary BYDI.

FIG. 6 is a diagram showing a configuration example of a field network.

FIG. 7 is a diagram showing a configuration example of a field likelihood BYUF, a field internal state BNAI, a field external input BGAI, and a field likelihood fixed flag BKOT.

FIG. 8 is a diagram showing a configuration example of a field weight BWAI.

FIG. 9 is a diagram showing a configuration example of a field designation screen.

FIG. 10 is a flowchart showing the operation of the character processing device.

FIG. 11 is a flowchart showing an operation of kana-kanji conversion.

FIG. 12 is a flowchart showing an operation of reading conversion.

FIG. 13 is a flowchart showing an operation of determining a break.

FIG. 14 is a flowchart showing an operation of determining a break in a field.

FIG. 15 is a flowchart showing an operation of obtaining a phrase likelihood according to a field.

FIG. 16 is a flowchart showing an operation of determining a conversion result.

FIG. 17 is a flowchart showing an operation of determining a converted word.

FIG. 18 is a flowchart showing an operation of determining a word to be converted in a field.

FIG. 19 is a flowchart showing an operation of updating a field external input by reading.

FIG. 20 is a flowchart showing an operation of correcting a conversion result.

FIG. 21 is a flowchart showing an operation of homonym display.

FIG. 22 is a flowchart showing the operation of changing the display order.

FIG. 23 is a flowchart showing an operation of updating an external input.

FIG. 24 is a flowchart showing an operation for setting a field environment.

[Explanation of symbols]

 CPU Microprocessor AB Address bus CB Control bus DB Data bus ROM Read only memory NK Program area PA Program area RAM Random access memory WDIC Word dictionary YDIC example dictionary BDIC field dictionary BYDI field reading dictionary BYUD field likelihood BNAI field Internal state BGAI field External Input BWAI field weight BKOT field likelihood fixed flag BMOD field mode WLIS latest word list DISK external memory KB keyboard KID conversion start instruction key SET environment setting activation key JUS address book creation function activation key KET candidate determination instruction key CHG field designation key CLR Field likelihood clear key MOD Field mode switching key CR Cursor register DBUF Display buffer memory CRTC CRT controller CRT display device CG character generator

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Yuji Kobayashi 3-30-2 Shimomaruko, Ota-ku, Tokyo Canon Inc. (72) Inventor Katsumi Masaki 3-30-2 Shimomaruko, Ota-ku, Tokyo Canon Within the corporation

Claims (5)

[Claims] 1. A kana-kanji conversion process for converting an input kana character string into a kanji-kana mixed sentence by referring to a dictionary in which the pronunciation of a word and the information of the word including the notation are stored in association with each other. In the character processing method to be executed, a plurality of groups are defined for the word group stored in the dictionary, and a group likelihood indicating the likelihood of appearance of words belonging to the group at the time of document input is set for each group. The group likelihood corresponding to at least one of the groups is updated according to the group to which the word converted by the kana-kanji conversion processing belongs, and in the kana-kanji conversion processing, the phrase for the kana character string to be converted is stored. From the candidates of the structure of, by comparing the group likelihood corresponding to the group to which the words constituting the candidate belong,
A character processing method characterized by determining the structure of a phrase to be converted. 2. The degree of association between the plurality of groups is stored, and in updating the group likelihood, the group likelihoods of groups other than the group to which the converted word belongs are updated based on the degree. The character processing method according to claim 1, which is updated. 3. The character processing method according to claim 2, wherein one or more desired groups are designated from the plurality of groups so that they can be excluded from the update targets of the group likelihood. 4. The character processing method according to claim 1, wherein the structure of the determined phrase includes a delimiter position of the phrase. 5. A dictionary in which word reading and word information including notation are stored in association with each other, input means for inputting characters, and kana character strings input by the input means are stored in the dictionary. Kana-kanji conversion means for converting to kanji-kana mixed sentences, group information storage means for storing a group to which a word belongs, and easiness of appearance of words belonging to the group as a group likelihood. Group likelihood storing means for storing each group, and group likelihood updating means for updating the group likelihood corresponding to at least one of the groups according to the group to which the converted word belongs by the kana-kanji converting means. And among the candidates of the structure of the bunsetsu for the kana character string to be converted by the kana-kanji conversion means, the phrase to which the candidate belongs belongs. Group likelihoods corresponding to
A character processing device comprising: a conversion control unit that determines a structure of a phrase to be converted.